@earendil-works/pi-ai 1.5.0 → 1.5.1

package/dist/api/openai-completions.js

@@ -555,12 +555,16 @@ function buildParams(model, context, options, compat = getCompat(model), cacheRe
         grammarToolInputProperties,
     });
     const cacheControl = getCompatCacheControl(compat, cacheRetention);
+    const promptCacheKeyMode = compat.promptCacheKeyMode ?? "auto";
+    const supportsPromptCacheKey = promptCacheKeyMode === "enabled" ||
+        (promptCacheKeyMode === "auto" &&
+            (model.baseUrl.includes("api.openai.com") ||
+                (cacheRetention === "long" && compat.supportsLongCacheRetention)));
     const params = {
         model: model.id,
         messages,
         stream: true,
-        prompt_cache_key: (model.baseUrl.includes("api.openai.com") && cacheRetention !== "none") ||
-            (cacheRetention === "long" && compat.supportsLongCacheRetention)
+        prompt_cache_key: cacheRetention !== "none" && supportsPromptCacheKey
             ? clampOpenAIPromptCacheKey(options?.sessionId)
             : undefined,
         prompt_cache_retention: cacheRetention === "long" && compat.supportsLongCacheRetention ? "24h" : undefined,
@@ -589,8 +593,7 @@ function buildParams(model, context, options, compat = getCompat(model), cacheRe
             params.tool_stream = true;
         }
     }
-    else if (hasToolHistory(context.messages)) {
-        // Anthropic (via LiteLLM/proxy) requires tools param when conversation has tool_calls/tool_results
+    else if (compat.requiresToolsWithToolHistory && hasToolHistory(context.messages)) {
         params.tools = [];
     }
     if (cacheControl) {
@@ -950,7 +953,7 @@ export function convertMessages(model, context, compat, options) {
             // Some providers don't accept null content, use empty string instead
             const assistantMsg = {
                 role: "assistant",
-                content: compat.requiresAssistantAfterToolResult ? "" : null,
+                content: compat.requiresAssistantAfterToolResult || compat.requiresNonNullAssistantContent ? "" : null,
             };
             const assistantTextParts = msg.content
                 .filter(isTextContentBlock)
@@ -1264,6 +1267,8 @@ function detectCompat(model) {
         maxTokensField: useMaxTokens ? "max_tokens" : "max_completion_tokens",
         requiresToolResultName: false,
         requiresAssistantAfterToolResult: false,
+        requiresNonNullAssistantContent: false,
+        requiresToolsWithToolHistory: false,
         requiresThinkingAsText: false,
         requiresReasoningContentOnAssistantMessages: isDeepSeek,
         thinkingFormat: isDeepSeek
@@ -1316,6 +1321,8 @@ function getCompat(model) {
         maxTokensField: model.compat.maxTokensField ?? detected.maxTokensField,
         requiresToolResultName: model.compat.requiresToolResultName ?? detected.requiresToolResultName,
         requiresAssistantAfterToolResult: model.compat.requiresAssistantAfterToolResult ?? detected.requiresAssistantAfterToolResult,
+        requiresNonNullAssistantContent: model.compat.requiresNonNullAssistantContent ?? detected.requiresNonNullAssistantContent,
+        requiresToolsWithToolHistory: model.compat.requiresToolsWithToolHistory ?? detected.requiresToolsWithToolHistory,
         requiresThinkingAsText: model.compat.requiresThinkingAsText ?? detected.requiresThinkingAsText,
         requiresReasoningContentOnAssistantMessages: model.compat.requiresReasoningContentOnAssistantMessages ??
             detected.requiresReasoningContentOnAssistantMessages,
@@ -1335,6 +1342,7 @@ function getCompat(model) {
         sendSessionAffinityHeaders: model.compat.sendSessionAffinityHeaders ?? detected.sendSessionAffinityHeaders,
         sessionAffinityFormat: model.compat.sessionAffinityFormat ?? detected.sessionAffinityFormat,
         supportsLongCacheRetention: model.compat.supportsLongCacheRetention ?? detected.supportsLongCacheRetention,
+        promptCacheKeyMode: model.compat.promptCacheKeyMode,
         vllmPriority: model.compat.vllmPriority,
     };
 }